Lift-and-Shift to Cloud NAS with 400 TB and Latency-Sensitive CAD
A design firm has 400 TB on a 5-year-old NAS. Of that, 250 TB is completed project archives untouched for more than 12 months, 100 TB is active project files, and 50 TB is rendered marketing media. About 120 engineers access active files over SMB with latency-sensitive CAD; archive retrieval averages 20 requests per month. The NAS is 85% full and growing about 6 TB per month. Cloud object storage costs $0.023/GB-month for standard and $0.004/GB-month for archive; egress is $0.09/GB. The office has a 1 Gbps internet link and a 500 Mbps dedicated cloud connection. The CIO wants to lift and shift everything to a cloud NAS.
1. Evaluate the CIO's lift-and-shift proposal. Which workloads should not move?
Read the worked answer
Lift-and-shift is the wrong first move. The active 100 TB is accessed by 120 engineers over SMB with latency-sensitive CAD; SMB metadata operations over a 500 Mbps cloud connect will add latency to every open, save, and file lock. The initial 400 TB copy is also impractical: 400 TB at 500 Mbps is 400 x 10^12 x 8 / 500 x 10^6 = 6.4 million seconds, about 74 days, ignoring overhead. A 1 Gbps internet link does not fix that for a shared office.
Do not move: active CAD project files, working renders, and anything with file locking or small random writes. Do move: the 250 TB archive untouched for more than 12 months. Consider moving completed renders only if distribution to clients is the main use, because egress will be charged on download.
Use a tiering gateway or policy engine that leaves a stub and recalls on access, seed the archive with a physical transfer appliance, and keep a local cache for the 20 monthly recalls. Re-evaluate active workload migration only after a pilot with 10 engineers and measured latency.
2. Propose a tiered data management plan and calculate rough monthly cloud storage and egress cost for the archive scenario.
Read the worked answer
Tier by access pattern, not by age alone.
- Archive: 250 TB to archive/object tier. At $0.004/GB-month: 250,000 GB x $0.004 = $1,000/month, or $12,000/year, before retrieval and request fees.
- Rendered media: 50 TB to standard object storage only if clients download it; at $0.023/GB-month that is 50,000 x $0.023 = $1,150/month. Egress at 2 TB/month = 2,000 x $0.09 = $180/month. Total roughly $2,330/month plus requests.
- Active 100 TB: keep on-premises NAS refreshed or expand it; do not put it in cloud NAS at this latency.
Add lifecycle rules to move projects after 12 months of no access, retention holds for legal, and an index so recalls are findable. Budget for retrieval fees on archive tier and for a cache. The CIO's 'everything' plan would cost at least 400,000 x $0.023 = $9,200/month for standard storage alone and still fail on latency and initial transfer time.
3. What network and data-movement changes are needed to make archive recall acceptable?
Read the worked answer
For 20 recalls per month, 2 TB each, the math matters: 2 TB at 500 Mbps is 2 x 10^12 x 8 / 500 x 10^6 = 32,000 seconds, about 8.9 hours; at 1 Gbps it is about 4.4 hours. That may be acceptable for archive but not for interactive work. Use the dedicated cloud connect for archive traffic, enable parallel multi-part transfer, and keep a local cache of the last 10-20 recalled projects. Compress and dedupe where the data type allows; image and video files may not compress much. Seed the initial 250 TB with an appliance rather than the wire. Monitor recall latency and egress monthly. If recalls become frequent, move that project back on-prem or to a cloud cache close to users; the tiering policy should be reversible. Also test a restore from archive before deleting the on-prem copy.